Forecasting with Dynamic Regression
Dynamic regression is a regression model that includes explanatory variables and/or dependent variables with time lags.
The dynamic regression approach therefore allows the development of models that combine the dynamics of time series type models and that allow explanatory variables or advanced indicators to be taken into account. The classic regression model is improved by the possibility of taking into account an autoregressive error term (Cochrane-Orcutt type), and by the use of lagged explanatory variables.
Complex method but …….
Dynamic regression can be used when (1) the data is long enough and stable enough to support a correlation model (2) the introduction of explanatory variables results in a significant improvement in model accuracy and (3) valid predictions are available for the explanatory variables. If some companies, which have the means to use dynamic regression models, are sometimes reluctant to use this modeling approach, it is because they know from experience that complex models often produce forecasts that are less good. than simpler models, even though they fit the observed data better.
Unlike univariate time series forecasting models (exponential smoothing, Box & Jenkins, ….), dynamic regression is a forecasting technique that allows highlighting the dynamic and causal effects between explanatory and lagged variables in the weather.
However, like the Box-Jenkins model, the dynamic regression model has a certain level of sophistication and has so far been difficult to identify and very time-consuming to construct. This has reduced its use in business forecasting.
….. but now easely implementable thanks to automation algorithms
Automation algorithms are now included in several forecasting software, in particular in the forecasting software “Forecast Pro”. They now allow forecasters to quickly and more easily build adaptable and more easily operational dynamic regression models. This has resulted in a greater use of this type of models.
Une description détaillée
A detailed description of the Dymanic Regression methodology is available in French here The english version will be available soon.
PREDICONSULT proposes a training course on “Forecasting Techniques“ that includes a detailed presentation of Exponential Smoothing.